will do this if this function is called by anyone other than the root
user.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9804 a95241bf-73f2-0310-859d-f6bbb57e9c96
for temporary memory anymore, it's the store's responsibility to do that
correctly now, and that functionality is reached via the vm_cache using
vm_cache_set_minimal_commitment().
Therefore, the vm_store commit() function now returns a status instead of
the size that could be commited.
Replaced the max_commit mechanism with one that cares about the available
memory, stores can reserve and unreserve such memory. The anonymous_commit()
will now fail in case it could not reserve the needed amount of memory.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9777 a95241bf-73f2-0310-859d-f6bbb57e9c96
with an empty queue, it crashed. sem_info::latest_holder is obviously not
set correctly, and should be fixed.
Some cosmetics.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9771 a95241bf-73f2-0310-859d-f6bbb57e9c96
in the wrong direction for some time now, corrupting the argument area (user
apps would crash).
B_SYSTEM_TEAM is now reserved, so that no team can be created there.
Fixed some warnings when debug output is enabled.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9719 a95241bf-73f2-0310-859d-f6bbb57e9c96
The inode now keeps track of the number of readers/writers.
Inode::BytesInChain() is now only used protected by the request lock.
Implemented some missing POSIX demands:
- when the last writer is closed, all waiting read requests are aborted
- pipefs_write() now returns EPIPE and signals the current thread with
SIGPIPE in case there are no readers left
- pipefs_read() will now return 0 in case there are no writers and no
unread buffers left in the pipe.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9717 a95241bf-73f2-0310-859d-f6bbb57e9c96
Correctly implemented devfs_create() - it will now open the an already
existing file if O_EXCL is not set, and no longer fail with EROFS.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9716 a95241bf-73f2-0310-859d-f6bbb57e9c96
divided into several parts, it could happen that overwrite the whole data
portion with data beyond the part that should have been read.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9706 a95241bf-73f2-0310-859d-f6bbb57e9c96
absolutely no reason to move them to B_SYSTEM_TEAM (additionally, during
startup, set_sem_owner() may crash).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9693 a95241bf-73f2-0310-859d-f6bbb57e9c96
Changed the last argument of _kern_mount() to be a string rather than a void pointer.
Greatly reduced the stack usage of _user_mount(); it now uses KPath instead
of on stack paths. It now also copies the parameter argument on the heap.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9692 a95241bf-73f2-0310-859d-f6bbb57e9c96
devfs_read_link() and rootfs_read_link() could write beyond the buffer
size passed in (off by one).
The comment for _kern_read_link() didn't fit to what the function really
did.
common_read_link() no longer null terminates the link - it's the file
system's responsibility to do that.
fs_read_link() is not supposed to return the length of the link anymore,
but only B_OK for success.
Note, we deviate slightly from POSIX here, where even a buffer too small
would be filled, and no terminating null byte has to be written at all.
We always return an error in case the buffer is too small, and the link
is not partially copied into the buffer.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9652 a95241bf-73f2-0310-859d-f6bbb57e9c96
Made the FS internal API a bit more consistent.
Removed devfs_unlink(), and devfs_rename() (which was just returning EROFS
anyway) - the devfs now appears completely read-only to the user. All
changes are triggered by kernel internal APIs (most usually through either
the devfs itself (old style drivers), the disk_device_manager, or the
standard device_manager).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9645 a95241bf-73f2-0310-859d-f6bbb57e9c96
Also implemented the now needed devfs_read_link() function.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9644 a95241bf-73f2-0310-859d-f6bbb57e9c96
yet, though, and just return B_ERROR.
devfs_unpublish_partition() now also returns B_ERROR since it's not yet
implemented.
Refactored the device publishing code, so that implementing devfs_publish_file_device()
will become easy.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9639 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added resolve_mount_point_to_volume_root() which does what
resolve_volume_root_to_mount_point() did. IOW the latter one didn't
do what it advertised.
* Fixed dir_vnode_to_path(). Basically broken due to the broken
resolve_volume_root_to_mount_point(), but also compared potentially
unrelated vnode IDs (belonging to different volumes).
* Fixed get_dir_path_and_leaf(). It didn't deal correctly with paths
ending in '/' (including the root dir).
* fs_mount() does now accept a NULL fsName, getting the FS name from
the DDM in this case.
* fs_mount() now also supports mounting file images; it lets the DDM
create a file device for them.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9629 a95241bf-73f2-0310-859d-f6bbb57e9c96
done by (currently not existing) devfs functions.
* Overridden GetMediaStatus() and GetGeometry() to return useful data.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9627 a95241bf-73f2-0310-859d-f6bbb57e9c96
Moved the debug macros into the program flow, so that they don't look so
temporary :)
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9557 a95241bf-73f2-0310-859d-f6bbb57e9c96
* fs_mount() and fs_unmount() are now aware of partitions. A device to
be mounted is looked up by the DDM and the mount state and mount
cookie of the corresponding KPartition are updated accordingly.
Still missing is support for mounting files. Moreover the device path
needs to be normalized.
* Added RecursiveLockAutoLocker class to ease the `goto errN' pain a bit.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9555 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Made GetMediaStatus() a bit more robust. If the ioctl fails (e.g. if it
is not implemented as in Thomas' drivers), it gets the device geometry
and does not fail, if the device is not removable.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9546 a95241bf-73f2-0310-859d-f6bbb57e9c96
and that main2() cannot therefore just wait until interrupts become
enabled.
Now, the main2() thread is resumed after interrupts are turned on.
I am not yet sure if this is really necessary, but I have the feeling that it is.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9531 a95241bf-73f2-0310-859d-f6bbb57e9c96
Axel Dörfler